Based on Simon Willison’s post on this topic, I went and wasted a fair bit of time today (okay, not actually wasted) on Google’s NotebookLM, uploading various sets of documents, and auto-generating podcast “deep dive” episodes from them.
Like the examples cited in Willison’s post above, I also found that putting in any content about you (such as my about page, and recent media coverage) yields embarrassingly positive portrayals of you that are both weirdly wrong but also strangely accurate, while also having this kind of sheen of … seductive flattery… Like the AIs want you to get puffed up on yourself, and get “high on your own supply” so to speak.
I got tired of hearing that though, so I finally went with a take that strictly talks about my books as the deep dive topic, and leaves me as the “author” entirely out of it. There are six somewhat thematically related AI Lore books that I used as the basis for this, including the two latest ones: Namaste, My Dude, Uncel, Relaxatopia, Anxietopia, The Jealous Human, and Das Machina.
I find it rather hard personally to explain what my books are, and was pleasantly surprised that the AI podcast “hosts” basically get it pretty right, sometimes seemingly even scarily insightful. I did quite reach the level of triggering in them an existential crisis, like one Reddit user was able to do, but by the end of it, there’s a heavy mist that descends on you as you slip into the Uncanny Ravine listening to these two go on and on.
Here is the link up on Spotify (and embedded below), as a sequel to my first installment of This AI Life, where I used an AI voice from Eleven Labs to interview me with questions that I wrote.
Also, if anyone’s curious, I used Riverside.fm’s free transcription service here to get a text version, which I fed into ChatGPT to write the episode summary. It’s nothing special, but simplifies my process.
